Using SIATECCompress to discover repeated themes and sections in polyphonic music

Three variants of the SIATECCompress algorithm were submitted to the 2016 MIREX competition on Discovery of Repeated Themes and Sections. The three variants were those that performed best in terms of three-layer F1 score (TLF1), three-layer precision (TLP) and three-layer recall (TLR), respectively, when a large number of variants of SIATECCompress, COSIATEC and Forth’s algorithm
were run on the polyphonic version of the JKU Patterns Development Database (JKU-PDD). The variant optimised for TLF1 achieved an average TLF1 score of 0.490 over the five pieces in the database. The variant optimised for TLP achieved an average TLP score of 0.561 over the JKU-PDD. The variant optimised for TLR achieved an average TLR score of 0.583 over the JKU-PDD. The three variants have been implemented in Java.